xref: /openbmc/linux/arch/arm64/kernel/head.S (revision 86e281fc)
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Low-level CPU initialisation
4 * Based on arch/arm/kernel/head.S
5 *
6 * Copyright (C) 1994-2002 Russell King
7 * Copyright (C) 2003-2012 ARM Ltd.
8 * Authors:	Catalin Marinas <catalin.marinas@arm.com>
9 *		Will Deacon <will.deacon@arm.com>
10 */
11
12#include <linux/linkage.h>
13#include <linux/init.h>
14#include <linux/pgtable.h>
15
16#include <asm/asm_pointer_auth.h>
17#include <asm/assembler.h>
18#include <asm/boot.h>
19#include <asm/bug.h>
20#include <asm/ptrace.h>
21#include <asm/asm-offsets.h>
22#include <asm/cache.h>
23#include <asm/cputype.h>
24#include <asm/el2_setup.h>
25#include <asm/elf.h>
26#include <asm/image.h>
27#include <asm/kernel-pgtable.h>
28#include <asm/kvm_arm.h>
29#include <asm/memory.h>
30#include <asm/pgtable-hwdef.h>
31#include <asm/page.h>
32#include <asm/scs.h>
33#include <asm/smp.h>
34#include <asm/sysreg.h>
35#include <asm/thread_info.h>
36#include <asm/virt.h>
37
38#include "efi-header.S"
39
40#if (PAGE_OFFSET & 0x1fffff) != 0
41#error PAGE_OFFSET must be at least 2MB aligned
42#endif
43
44/*
45 * Kernel startup entry point.
46 * ---------------------------
47 *
48 * The requirements are:
49 *   MMU = off, D-cache = off, I-cache = on or off,
50 *   x0 = physical address to the FDT blob.
51 *
52 * Note that the callee-saved registers are used for storing variables
53 * that are useful before the MMU is enabled. The allocations are described
54 * in the entry routines.
55 */
56	__HEAD
57	/*
58	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
59	 */
60	efi_signature_nop			// special NOP to identity as PE/COFF executable
61	b	primary_entry			// branch to kernel start, magic
62	.quad	0				// Image load offset from start of RAM, little-endian
63	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
64	le64sym	_kernel_flags_le		// Informative flags, little-endian
65	.quad	0				// reserved
66	.quad	0				// reserved
67	.quad	0				// reserved
68	.ascii	ARM64_IMAGE_MAGIC		// Magic number
69	.long	.Lpe_header_offset		// Offset to the PE header.
70
71	__EFI_PE_HEADER
72
73	.section ".idmap.text","a"
74
75	/*
76	 * The following callee saved general purpose registers are used on the
77	 * primary lowlevel boot path:
78	 *
79	 *  Register   Scope                      Purpose
80	 *  x19        primary_entry() .. start_kernel()        whether we entered with the MMU on
81	 *  x20        primary_entry() .. __primary_switch()    CPU boot mode
82	 *  x21        primary_entry() .. start_kernel()        FDT pointer passed at boot in x0
83	 *  x22        create_idmap() .. start_kernel()         ID map VA of the DT blob
84	 *  x23        primary_entry() .. start_kernel()        physical misalignment/KASLR offset
85	 *  x24        __primary_switch()                       linear map KASLR seed
86	 *  x25        primary_entry() .. start_kernel()        supported VA size
87	 *  x28        create_idmap()                           callee preserved temp register
88	 */
89SYM_CODE_START(primary_entry)
90	bl	record_mmu_state
91	bl	preserve_boot_args
92	bl	create_idmap
93
94	/*
95	 * If we entered with the MMU and caches on, clean the ID mapped part
96	 * of the primary boot code to the PoC so we can safely execute it with
97	 * the MMU off.
98	 */
99	cbz	x19, 0f
100	adrp	x0, __idmap_text_start
101	adr_l	x1, __idmap_text_end
102	adr_l	x2, dcache_clean_poc
103	blr	x2
1040:	mov	x0, x19
105	bl	init_kernel_el			// w0=cpu_boot_mode
106	mov	x20, x0
107
108	/*
109	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
110	 * details.
111	 * On return, the CPU will be ready for the MMU to be turned on and
112	 * the TCR will have been set.
113	 */
114#if VA_BITS > 48
115	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
116	tst	x0, ID_AA64MMFR2_EL1_VARange_MASK
117	mov	x0, #VA_BITS
118	mov	x25, #VA_BITS_MIN
119	csel	x25, x25, x0, eq
120	mov	x0, x25
121#endif
122	bl	__cpu_setup			// initialise processor
123	b	__primary_switch
124SYM_CODE_END(primary_entry)
125
126	__INIT
127SYM_CODE_START_LOCAL(record_mmu_state)
128	mrs	x19, CurrentEL
129	cmp	x19, #CurrentEL_EL2
130	mrs	x19, sctlr_el1
131	b.ne	0f
132	mrs	x19, sctlr_el2
1330:
134CPU_LE( tbnz	x19, #SCTLR_ELx_EE_SHIFT, 1f	)
135CPU_BE( tbz	x19, #SCTLR_ELx_EE_SHIFT, 1f	)
136	tst	x19, #SCTLR_ELx_C		// Z := (C == 0)
137	and	x19, x19, #SCTLR_ELx_M		// isolate M bit
138	csel	x19, xzr, x19, eq		// clear x19 if Z
139	ret
140
141	/*
142	 * Set the correct endianness early so all memory accesses issued
143	 * before init_kernel_el() occur in the correct byte order. Note that
144	 * this means the MMU must be disabled, or the active ID map will end
145	 * up getting interpreted with the wrong byte order.
146	 */
1471:	eor	x19, x19, #SCTLR_ELx_EE
148	bic	x19, x19, #SCTLR_ELx_M
149	b.ne	2f
150	pre_disable_mmu_workaround
151	msr	sctlr_el2, x19
152	b	3f
1532:	pre_disable_mmu_workaround
154	msr	sctlr_el1, x19
1553:	isb
156	mov	x19, xzr
157	ret
158SYM_CODE_END(record_mmu_state)
159
160/*
161 * Preserve the arguments passed by the bootloader in x0 .. x3
162 */
163SYM_CODE_START_LOCAL(preserve_boot_args)
164	mov	x21, x0				// x21=FDT
165
166	adr_l	x0, boot_args			// record the contents of
167	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
168	stp	x2, x3, [x0, #16]
169
170	cbnz	x19, 0f				// skip cache invalidation if MMU is on
171	dmb	sy				// needed before dc ivac with
172						// MMU off
173
174	add	x1, x0, #0x20			// 4 x 8 bytes
175	b	dcache_inval_poc		// tail call
1760:	str_l   x19, mmu_enabled_at_boot, x0
177	ret
178SYM_CODE_END(preserve_boot_args)
179
180SYM_FUNC_START_LOCAL(clear_page_tables)
181	/*
182	 * Clear the init page tables.
183	 */
184	adrp	x0, init_pg_dir
185	adrp	x1, init_pg_end
186	sub	x2, x1, x0
187	mov	x1, xzr
188	b	__pi_memset			// tail call
189SYM_FUNC_END(clear_page_tables)
190
191/*
192 * Macro to populate page table entries, these entries can be pointers to the next level
193 * or last level entries pointing to physical memory.
194 *
195 *	tbl:	page table address
196 *	rtbl:	pointer to page table or physical memory
197 *	index:	start index to write
198 *	eindex:	end index to write - [index, eindex] written to
199 *	flags:	flags for pagetable entry to or in
200 *	inc:	increment to rtbl between each entry
201 *	tmp1:	temporary variable
202 *
203 * Preserves:	tbl, eindex, flags, inc
204 * Corrupts:	index, tmp1
205 * Returns:	rtbl
206 */
207	.macro populate_entries, tbl, rtbl, index, eindex, flags, inc, tmp1
208.Lpe\@:	phys_to_pte \tmp1, \rtbl
209	orr	\tmp1, \tmp1, \flags	// tmp1 = table entry
210	str	\tmp1, [\tbl, \index, lsl #3]
211	add	\rtbl, \rtbl, \inc	// rtbl = pa next level
212	add	\index, \index, #1
213	cmp	\index, \eindex
214	b.ls	.Lpe\@
215	.endm
216
217/*
218 * Compute indices of table entries from virtual address range. If multiple entries
219 * were needed in the previous page table level then the next page table level is assumed
220 * to be composed of multiple pages. (This effectively scales the end index).
221 *
222 *	vstart:	virtual address of start of range
223 *	vend:	virtual address of end of range - we map [vstart, vend]
224 *	shift:	shift used to transform virtual address into index
225 *	order:  #imm 2log(number of entries in page table)
226 *	istart:	index in table corresponding to vstart
227 *	iend:	index in table corresponding to vend
228 *	count:	On entry: how many extra entries were required in previous level, scales
229 *			  our end index.
230 *		On exit: returns how many extra entries required for next page table level
231 *
232 * Preserves:	vstart, vend
233 * Returns:	istart, iend, count
234 */
235	.macro compute_indices, vstart, vend, shift, order, istart, iend, count
236	ubfx	\istart, \vstart, \shift, \order
237	ubfx	\iend, \vend, \shift, \order
238	add	\iend, \iend, \count, lsl \order
239	sub	\count, \iend, \istart
240	.endm
241
242/*
243 * Map memory for specified virtual address range. Each level of page table needed supports
244 * multiple entries. If a level requires n entries the next page table level is assumed to be
245 * formed from n pages.
246 *
247 *	tbl:	location of page table
248 *	rtbl:	address to be used for first level page table entry (typically tbl + PAGE_SIZE)
249 *	vstart:	virtual address of start of range
250 *	vend:	virtual address of end of range - we map [vstart, vend - 1]
251 *	flags:	flags to use to map last level entries
252 *	phys:	physical address corresponding to vstart - physical memory is contiguous
253 *	order:  #imm 2log(number of entries in PGD table)
254 *
255 * If extra_shift is set, an extra level will be populated if the end address does
256 * not fit in 'extra_shift' bits. This assumes vend is in the TTBR0 range.
257 *
258 * Temporaries:	istart, iend, tmp, count, sv - these need to be different registers
259 * Preserves:	vstart, flags
260 * Corrupts:	tbl, rtbl, vend, istart, iend, tmp, count, sv
261 */
262	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, order, istart, iend, tmp, count, sv, extra_shift
263	sub \vend, \vend, #1
264	add \rtbl, \tbl, #PAGE_SIZE
265	mov \count, #0
266
267	.ifnb	\extra_shift
268	tst	\vend, #~((1 << (\extra_shift)) - 1)
269	b.eq	.L_\@
270	compute_indices \vstart, \vend, #\extra_shift, #(PAGE_SHIFT - 3), \istart, \iend, \count
271	mov \sv, \rtbl
272	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
273	mov \tbl, \sv
274	.endif
275.L_\@:
276	compute_indices \vstart, \vend, #PGDIR_SHIFT, #\order, \istart, \iend, \count
277	mov \sv, \rtbl
278	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
279	mov \tbl, \sv
280
281#if SWAPPER_PGTABLE_LEVELS > 3
282	compute_indices \vstart, \vend, #PUD_SHIFT, #(PAGE_SHIFT - 3), \istart, \iend, \count
283	mov \sv, \rtbl
284	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
285	mov \tbl, \sv
286#endif
287
288#if SWAPPER_PGTABLE_LEVELS > 2
289	compute_indices \vstart, \vend, #SWAPPER_TABLE_SHIFT, #(PAGE_SHIFT - 3), \istart, \iend, \count
290	mov \sv, \rtbl
291	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
292	mov \tbl, \sv
293#endif
294
295	compute_indices \vstart, \vend, #SWAPPER_BLOCK_SHIFT, #(PAGE_SHIFT - 3), \istart, \iend, \count
296	bic \rtbl, \phys, #SWAPPER_BLOCK_SIZE - 1
297	populate_entries \tbl, \rtbl, \istart, \iend, \flags, #SWAPPER_BLOCK_SIZE, \tmp
298	.endm
299
300/*
301 * Remap a subregion created with the map_memory macro with modified attributes
302 * or output address. The entire remapped region must have been covered in the
303 * invocation of map_memory.
304 *
305 * x0: last level table address (returned in first argument to map_memory)
306 * x1: start VA of the existing mapping
307 * x2: start VA of the region to update
308 * x3: end VA of the region to update (exclusive)
309 * x4: start PA associated with the region to update
310 * x5: attributes to set on the updated region
311 * x6: order of the last level mappings
312 */
313SYM_FUNC_START_LOCAL(remap_region)
314	sub	x3, x3, #1		// make end inclusive
315
316	// Get the index offset for the start of the last level table
317	lsr	x1, x1, x6
318	bfi	x1, xzr, #0, #PAGE_SHIFT - 3
319
320	// Derive the start and end indexes into the last level table
321	// associated with the provided region
322	lsr	x2, x2, x6
323	lsr	x3, x3, x6
324	sub	x2, x2, x1
325	sub	x3, x3, x1
326
327	mov	x1, #1
328	lsl	x6, x1, x6		// block size at this level
329
330	populate_entries x0, x4, x2, x3, x5, x6, x7
331	ret
332SYM_FUNC_END(remap_region)
333
334SYM_FUNC_START_LOCAL(create_idmap)
335	mov	x28, lr
336	/*
337	 * The ID map carries a 1:1 mapping of the physical address range
338	 * covered by the loaded image, which could be anywhere in DRAM. This
339	 * means that the required size of the VA (== PA) space is decided at
340	 * boot time, and could be more than the configured size of the VA
341	 * space for ordinary kernel and user space mappings.
342	 *
343	 * There are three cases to consider here:
344	 * - 39 <= VA_BITS < 48, and the ID map needs up to 48 VA bits to cover
345	 *   the placement of the image. In this case, we configure one extra
346	 *   level of translation on the fly for the ID map only. (This case
347	 *   also covers 42-bit VA/52-bit PA on 64k pages).
348	 *
349	 * - VA_BITS == 48, and the ID map needs more than 48 VA bits. This can
350	 *   only happen when using 64k pages, in which case we need to extend
351	 *   the root level table rather than add a level. Note that we can
352	 *   treat this case as 'always extended' as long as we take care not
353	 *   to program an unsupported T0SZ value into the TCR register.
354	 *
355	 * - Combinations that would require two additional levels of
356	 *   translation are not supported, e.g., VA_BITS==36 on 16k pages, or
357	 *   VA_BITS==39/4k pages with 5-level paging, where the input address
358	 *   requires more than 47 or 48 bits, respectively.
359	 */
360#if (VA_BITS < 48)
361#define IDMAP_PGD_ORDER	(VA_BITS - PGDIR_SHIFT)
362#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
363
364	/*
365	 * If VA_BITS < 48, we have to configure an additional table level.
366	 * First, we have to verify our assumption that the current value of
367	 * VA_BITS was chosen such that all translation levels are fully
368	 * utilised, and that lowering T0SZ will always result in an additional
369	 * translation level to be configured.
370	 */
371#if VA_BITS != EXTRA_SHIFT
372#error "Mismatch between VA_BITS and page size/number of translation levels"
373#endif
374#else
375#define IDMAP_PGD_ORDER	(PHYS_MASK_SHIFT - PGDIR_SHIFT)
376#define EXTRA_SHIFT
377	/*
378	 * If VA_BITS == 48, we don't have to configure an additional
379	 * translation level, but the top-level table has more entries.
380	 */
381#endif
382	adrp	x0, init_idmap_pg_dir
383	adrp	x3, _text
384	adrp	x6, _end + MAX_FDT_SIZE + SWAPPER_BLOCK_SIZE
385	mov_q	x7, SWAPPER_RX_MMUFLAGS
386
387	map_memory x0, x1, x3, x6, x7, x3, IDMAP_PGD_ORDER, x10, x11, x12, x13, x14, EXTRA_SHIFT
388
389	/* Remap the kernel page tables r/w in the ID map */
390	adrp	x1, _text
391	adrp	x2, init_pg_dir
392	adrp	x3, init_pg_end
393	bic	x4, x2, #SWAPPER_BLOCK_SIZE - 1
394	mov_q	x5, SWAPPER_RW_MMUFLAGS
395	mov	x6, #SWAPPER_BLOCK_SHIFT
396	bl	remap_region
397
398	/* Remap the FDT after the kernel image */
399	adrp	x1, _text
400	adrp	x22, _end + SWAPPER_BLOCK_SIZE
401	bic	x2, x22, #SWAPPER_BLOCK_SIZE - 1
402	bfi	x22, x21, #0, #SWAPPER_BLOCK_SHIFT		// remapped FDT address
403	add	x3, x2, #MAX_FDT_SIZE + SWAPPER_BLOCK_SIZE
404	bic	x4, x21, #SWAPPER_BLOCK_SIZE - 1
405	mov_q	x5, SWAPPER_RW_MMUFLAGS
406	mov	x6, #SWAPPER_BLOCK_SHIFT
407	bl	remap_region
408
409	/*
410	 * Since the page tables have been populated with non-cacheable
411	 * accesses (MMU disabled), invalidate those tables again to
412	 * remove any speculatively loaded cache lines.
413	 */
414	cbnz	x19, 0f				// skip cache invalidation if MMU is on
415	dmb	sy
416
417	adrp	x0, init_idmap_pg_dir
418	adrp	x1, init_idmap_pg_end
419	bl	dcache_inval_poc
4200:	ret	x28
421SYM_FUNC_END(create_idmap)
422
423SYM_FUNC_START_LOCAL(create_kernel_mapping)
424	adrp	x0, init_pg_dir
425	mov_q	x5, KIMAGE_VADDR		// compile time __va(_text)
426#ifdef CONFIG_RELOCATABLE
427	add	x5, x5, x23			// add KASLR displacement
428#endif
429	adrp	x6, _end			// runtime __pa(_end)
430	adrp	x3, _text			// runtime __pa(_text)
431	sub	x6, x6, x3			// _end - _text
432	add	x6, x6, x5			// runtime __va(_end)
433	mov_q	x7, SWAPPER_RW_MMUFLAGS
434
435	map_memory x0, x1, x5, x6, x7, x3, (VA_BITS - PGDIR_SHIFT), x10, x11, x12, x13, x14
436
437	dsb	ishst				// sync with page table walker
438	ret
439SYM_FUNC_END(create_kernel_mapping)
440
441	/*
442	 * Initialize CPU registers with task-specific and cpu-specific context.
443	 *
444	 * Create a final frame record at task_pt_regs(current)->stackframe, so
445	 * that the unwinder can identify the final frame record of any task by
446	 * its location in the task stack. We reserve the entire pt_regs space
447	 * for consistency with user tasks and kthreads.
448	 */
449	.macro	init_cpu_task tsk, tmp1, tmp2
450	msr	sp_el0, \tsk
451
452	ldr	\tmp1, [\tsk, #TSK_STACK]
453	add	sp, \tmp1, #THREAD_SIZE
454	sub	sp, sp, #PT_REGS_SIZE
455
456	stp	xzr, xzr, [sp, #S_STACKFRAME]
457	add	x29, sp, #S_STACKFRAME
458
459	scs_load_current
460
461	adr_l	\tmp1, __per_cpu_offset
462	ldr	w\tmp2, [\tsk, #TSK_TI_CPU]
463	ldr	\tmp1, [\tmp1, \tmp2, lsl #3]
464	set_this_cpu_offset \tmp1
465	.endm
466
467/*
468 * The following fragment of code is executed with the MMU enabled.
469 *
470 *   x0 = __pa(KERNEL_START)
471 */
472SYM_FUNC_START_LOCAL(__primary_switched)
473	adr_l	x4, init_task
474	init_cpu_task x4, x5, x6
475
476	adr_l	x8, vectors			// load VBAR_EL1 with virtual
477	msr	vbar_el1, x8			// vector table address
478	isb
479
480	stp	x29, x30, [sp, #-16]!
481	mov	x29, sp
482
483	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
484
485	ldr_l	x4, kimage_vaddr		// Save the offset between
486	sub	x4, x4, x0			// the kernel virtual and
487	str_l	x4, kimage_voffset, x5		// physical mappings
488
489	mov	x0, x20
490	bl	set_cpu_boot_mode_flag
491
492	// Clear BSS
493	adr_l	x0, __bss_start
494	mov	x1, xzr
495	adr_l	x2, __bss_stop
496	sub	x2, x2, x0
497	bl	__pi_memset
498	dsb	ishst				// Make zero page visible to PTW
499
500#if VA_BITS > 48
501	adr_l	x8, vabits_actual		// Set this early so KASAN early init
502	str	x25, [x8]			// ... observes the correct value
503	dc	civac, x8			// Make visible to booting secondaries
504#endif
505
506#ifdef CONFIG_RANDOMIZE_BASE
507	adrp	x5, memstart_offset_seed	// Save KASLR linear map seed
508	strh	w24, [x5, :lo12:memstart_offset_seed]
509#endif
510#if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
511	bl	kasan_early_init
512#endif
513	mov	x0, x21				// pass FDT address in x0
514	bl	early_fdt_map			// Try mapping the FDT early
515	mov	x0, x20				// pass the full boot status
516	bl	init_feature_override		// Parse cpu feature overrides
517#ifdef CONFIG_UNWIND_PATCH_PAC_INTO_SCS
518	bl	scs_patch_vmlinux
519#endif
520	mov	x0, x20
521	bl	finalise_el2			// Prefer VHE if possible
522	ldp	x29, x30, [sp], #16
523	bl	start_kernel
524	ASM_BUG()
525SYM_FUNC_END(__primary_switched)
526
527/*
528 * end early head section, begin head code that is also used for
529 * hotplug and needs to have the same protections as the text region
530 */
531	.section ".idmap.text","a"
532
533/*
534 * Starting from EL2 or EL1, configure the CPU to execute at the highest
535 * reachable EL supported by the kernel in a chosen default state. If dropping
536 * from EL2 to EL1, configure EL2 before configuring EL1.
537 *
538 * Since we cannot always rely on ERET synchronizing writes to sysregs (e.g. if
539 * SCTLR_ELx.EOS is clear), we place an ISB prior to ERET.
540 *
541 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in x0 if
542 * booted in EL1 or EL2 respectively, with the top 32 bits containing
543 * potential context flags. These flags are *not* stored in __boot_cpu_mode.
544 *
545 * x0: whether we are being called from the primary boot path with the MMU on
546 */
547SYM_FUNC_START(init_kernel_el)
548	mrs	x1, CurrentEL
549	cmp	x1, #CurrentEL_EL2
550	b.eq	init_el2
551
552SYM_INNER_LABEL(init_el1, SYM_L_LOCAL)
553	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
554	pre_disable_mmu_workaround
555	msr	sctlr_el1, x0
556	isb
557	mov_q	x0, INIT_PSTATE_EL1
558	msr	spsr_el1, x0
559	msr	elr_el1, lr
560	mov	w0, #BOOT_CPU_MODE_EL1
561	eret
562
563SYM_INNER_LABEL(init_el2, SYM_L_LOCAL)
564	msr	elr_el2, lr
565
566	// clean all HYP code to the PoC if we booted at EL2 with the MMU on
567	cbz	x0, 0f
568	adrp	x0, __hyp_idmap_text_start
569	adr_l	x1, __hyp_text_end
570	adr_l	x2, dcache_clean_poc
571	blr	x2
5720:
573	mov_q	x0, HCR_HOST_NVHE_FLAGS
574	msr	hcr_el2, x0
575	isb
576
577	init_el2_state
578
579	/* Hypervisor stub */
580	adr_l	x0, __hyp_stub_vectors
581	msr	vbar_el2, x0
582	isb
583
584	mov_q	x1, INIT_SCTLR_EL1_MMU_OFF
585
586	/*
587	 * Fruity CPUs seem to have HCR_EL2.E2H set to RES1,
588	 * making it impossible to start in nVHE mode. Is that
589	 * compliant with the architecture? Absolutely not!
590	 */
591	mrs	x0, hcr_el2
592	and	x0, x0, #HCR_E2H
593	cbz	x0, 1f
594
595	/* Set a sane SCTLR_EL1, the VHE way */
596	pre_disable_mmu_workaround
597	msr_s	SYS_SCTLR_EL12, x1
598	mov	x2, #BOOT_CPU_FLAG_E2H
599	b	2f
600
6011:
602	pre_disable_mmu_workaround
603	msr	sctlr_el1, x1
604	mov	x2, xzr
6052:
606	__init_el2_nvhe_prepare_eret
607
608	mov	w0, #BOOT_CPU_MODE_EL2
609	orr	x0, x0, x2
610	eret
611SYM_FUNC_END(init_kernel_el)
612
613	/*
614	 * This provides a "holding pen" for platforms to hold all secondary
615	 * cores are held until we're ready for them to initialise.
616	 */
617SYM_FUNC_START(secondary_holding_pen)
618	mov	x0, xzr
619	bl	init_kernel_el			// w0=cpu_boot_mode
620	mrs	x2, mpidr_el1
621	mov_q	x1, MPIDR_HWID_BITMASK
622	and	x2, x2, x1
623	adr_l	x3, secondary_holding_pen_release
624pen:	ldr	x4, [x3]
625	cmp	x4, x2
626	b.eq	secondary_startup
627	wfe
628	b	pen
629SYM_FUNC_END(secondary_holding_pen)
630
631	/*
632	 * Secondary entry point that jumps straight into the kernel. Only to
633	 * be used where CPUs are brought online dynamically by the kernel.
634	 */
635SYM_FUNC_START(secondary_entry)
636	mov	x0, xzr
637	bl	init_kernel_el			// w0=cpu_boot_mode
638	b	secondary_startup
639SYM_FUNC_END(secondary_entry)
640
641SYM_FUNC_START_LOCAL(secondary_startup)
642	/*
643	 * Common entry point for secondary CPUs.
644	 */
645	mov	x20, x0				// preserve boot mode
646	bl	__cpu_secondary_check52bitva
647#if VA_BITS > 48
648	ldr_l	x0, vabits_actual
649#endif
650	bl	__cpu_setup			// initialise processor
651	adrp	x1, swapper_pg_dir
652	adrp	x2, idmap_pg_dir
653	bl	__enable_mmu
654	ldr	x8, =__secondary_switched
655	br	x8
656SYM_FUNC_END(secondary_startup)
657
658	.text
659SYM_FUNC_START_LOCAL(__secondary_switched)
660	mov	x0, x20
661	bl	set_cpu_boot_mode_flag
662
663	mov	x0, x20
664	bl	finalise_el2
665
666	str_l	xzr, __early_cpu_boot_status, x3
667	adr_l	x5, vectors
668	msr	vbar_el1, x5
669	isb
670
671	adr_l	x0, secondary_data
672	ldr	x2, [x0, #CPU_BOOT_TASK]
673	cbz	x2, __secondary_too_slow
674
675	init_cpu_task x2, x1, x3
676
677#ifdef CONFIG_ARM64_PTR_AUTH
678	ptrauth_keys_init_cpu x2, x3, x4, x5
679#endif
680
681	bl	secondary_start_kernel
682	ASM_BUG()
683SYM_FUNC_END(__secondary_switched)
684
685SYM_FUNC_START_LOCAL(__secondary_too_slow)
686	wfe
687	wfi
688	b	__secondary_too_slow
689SYM_FUNC_END(__secondary_too_slow)
690
691/*
692 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
693 * in w0. See arch/arm64/include/asm/virt.h for more info.
694 */
695SYM_FUNC_START_LOCAL(set_cpu_boot_mode_flag)
696	adr_l	x1, __boot_cpu_mode
697	cmp	w0, #BOOT_CPU_MODE_EL2
698	b.ne	1f
699	add	x1, x1, #4
7001:	str	w0, [x1]			// Save CPU boot mode
701	ret
702SYM_FUNC_END(set_cpu_boot_mode_flag)
703
704/*
705 * The booting CPU updates the failed status @__early_cpu_boot_status,
706 * with MMU turned off.
707 *
708 * update_early_cpu_boot_status tmp, status
709 *  - Corrupts tmp1, tmp2
710 *  - Writes 'status' to __early_cpu_boot_status and makes sure
711 *    it is committed to memory.
712 */
713
714	.macro	update_early_cpu_boot_status status, tmp1, tmp2
715	mov	\tmp2, #\status
716	adr_l	\tmp1, __early_cpu_boot_status
717	str	\tmp2, [\tmp1]
718	dmb	sy
719	dc	ivac, \tmp1			// Invalidate potentially stale cache line
720	.endm
721
722/*
723 * Enable the MMU.
724 *
725 *  x0  = SCTLR_EL1 value for turning on the MMU.
726 *  x1  = TTBR1_EL1 value
727 *  x2  = ID map root table address
728 *
729 * Returns to the caller via x30/lr. This requires the caller to be covered
730 * by the .idmap.text section.
731 *
732 * Checks if the selected granule size is supported by the CPU.
733 * If it isn't, park the CPU
734 */
735	.section ".idmap.text","a"
736SYM_FUNC_START(__enable_mmu)
737	mrs	x3, ID_AA64MMFR0_EL1
738	ubfx	x3, x3, #ID_AA64MMFR0_EL1_TGRAN_SHIFT, 4
739	cmp     x3, #ID_AA64MMFR0_EL1_TGRAN_SUPPORTED_MIN
740	b.lt    __no_granule_support
741	cmp     x3, #ID_AA64MMFR0_EL1_TGRAN_SUPPORTED_MAX
742	b.gt    __no_granule_support
743	phys_to_ttbr x2, x2
744	msr	ttbr0_el1, x2			// load TTBR0
745	load_ttbr1 x1, x1, x3
746
747	set_sctlr_el1	x0
748
749	ret
750SYM_FUNC_END(__enable_mmu)
751
752SYM_FUNC_START(__cpu_secondary_check52bitva)
753#if VA_BITS > 48
754	ldr_l	x0, vabits_actual
755	cmp	x0, #52
756	b.ne	2f
757
758	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
759	and	x0, x0, ID_AA64MMFR2_EL1_VARange_MASK
760	cbnz	x0, 2f
761
762	update_early_cpu_boot_status \
763		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
7641:	wfe
765	wfi
766	b	1b
767
768#endif
7692:	ret
770SYM_FUNC_END(__cpu_secondary_check52bitva)
771
772SYM_FUNC_START_LOCAL(__no_granule_support)
773	/* Indicate that this CPU can't boot and is stuck in the kernel */
774	update_early_cpu_boot_status \
775		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
7761:
777	wfe
778	wfi
779	b	1b
780SYM_FUNC_END(__no_granule_support)
781
782#ifdef CONFIG_RELOCATABLE
783SYM_FUNC_START_LOCAL(__relocate_kernel)
784	/*
785	 * Iterate over each entry in the relocation table, and apply the
786	 * relocations in place.
787	 */
788	adr_l	x9, __rela_start
789	adr_l	x10, __rela_end
790	mov_q	x11, KIMAGE_VADDR		// default virtual offset
791	add	x11, x11, x23			// actual virtual offset
792
7930:	cmp	x9, x10
794	b.hs	1f
795	ldp	x12, x13, [x9], #24
796	ldr	x14, [x9, #-8]
797	cmp	w13, #R_AARCH64_RELATIVE
798	b.ne	0b
799	add	x14, x14, x23			// relocate
800	str	x14, [x12, x23]
801	b	0b
802
8031:
804#ifdef CONFIG_RELR
805	/*
806	 * Apply RELR relocations.
807	 *
808	 * RELR is a compressed format for storing relative relocations. The
809	 * encoded sequence of entries looks like:
810	 * [ AAAAAAAA BBBBBBB1 BBBBBBB1 ... AAAAAAAA BBBBBB1 ... ]
811	 *
812	 * i.e. start with an address, followed by any number of bitmaps. The
813	 * address entry encodes 1 relocation. The subsequent bitmap entries
814	 * encode up to 63 relocations each, at subsequent offsets following
815	 * the last address entry.
816	 *
817	 * The bitmap entries must have 1 in the least significant bit. The
818	 * assumption here is that an address cannot have 1 in lsb. Odd
819	 * addresses are not supported. Any odd addresses are stored in the RELA
820	 * section, which is handled above.
821	 *
822	 * Excluding the least significant bit in the bitmap, each non-zero
823	 * bit in the bitmap represents a relocation to be applied to
824	 * a corresponding machine word that follows the base address
825	 * word. The second least significant bit represents the machine
826	 * word immediately following the initial address, and each bit
827	 * that follows represents the next word, in linear order. As such,
828	 * a single bitmap can encode up to 63 relocations in a 64-bit object.
829	 *
830	 * In this implementation we store the address of the next RELR table
831	 * entry in x9, the address being relocated by the current address or
832	 * bitmap entry in x13 and the address being relocated by the current
833	 * bit in x14.
834	 */
835	adr_l	x9, __relr_start
836	adr_l	x10, __relr_end
837
8382:	cmp	x9, x10
839	b.hs	7f
840	ldr	x11, [x9], #8
841	tbnz	x11, #0, 3f			// branch to handle bitmaps
842	add	x13, x11, x23
843	ldr	x12, [x13]			// relocate address entry
844	add	x12, x12, x23
845	str	x12, [x13], #8			// adjust to start of bitmap
846	b	2b
847
8483:	mov	x14, x13
8494:	lsr	x11, x11, #1
850	cbz	x11, 6f
851	tbz	x11, #0, 5f			// skip bit if not set
852	ldr	x12, [x14]			// relocate bit
853	add	x12, x12, x23
854	str	x12, [x14]
855
8565:	add	x14, x14, #8			// move to next bit's address
857	b	4b
858
8596:	/*
860	 * Move to the next bitmap's address. 8 is the word size, and 63 is the
861	 * number of significant bits in a bitmap entry.
862	 */
863	add	x13, x13, #(8 * 63)
864	b	2b
865
8667:
867#endif
868	ret
869
870SYM_FUNC_END(__relocate_kernel)
871#endif
872
873SYM_FUNC_START_LOCAL(__primary_switch)
874	adrp	x1, reserved_pg_dir
875	adrp	x2, init_idmap_pg_dir
876	bl	__enable_mmu
877#ifdef CONFIG_RELOCATABLE
878	adrp	x23, KERNEL_START
879	and	x23, x23, MIN_KIMG_ALIGN - 1
880#ifdef CONFIG_RANDOMIZE_BASE
881	mov	x0, x22
882	adrp	x1, init_pg_end
883	mov	sp, x1
884	mov	x29, xzr
885	bl	__pi_kaslr_early_init
886	and	x24, x0, #SZ_2M - 1		// capture memstart offset seed
887	bic	x0, x0, #SZ_2M - 1
888	orr	x23, x23, x0			// record kernel offset
889#endif
890#endif
891	bl	clear_page_tables
892	bl	create_kernel_mapping
893
894	adrp	x1, init_pg_dir
895	load_ttbr1 x1, x1, x2
896#ifdef CONFIG_RELOCATABLE
897	bl	__relocate_kernel
898#endif
899	ldr	x8, =__primary_switched
900	adrp	x0, KERNEL_START		// __pa(KERNEL_START)
901	br	x8
902SYM_FUNC_END(__primary_switch)
903